[edk2-devel] [PATCH v4 00/11] ArmPkg: implement EFI memory attributes protocol

[Ard Biesheuvel]

On Wed, 8 Mar 2023 at 18:24, Taylor Beebe <t at taylorbeebe.com> wrote:
>
> My mistake - the DEBUG_PROPERTY_DEBUG_CLEAR_MEMORY_ENABLED feature is
> why FreePagesWithProtectionAttributesTestCase might fail.
>
> To make the clear memory feature more compatible with the memory
> attribute protocol, can you add a check to DebugClearMemoryEnabled() on
> free and clear EFI_MEMORY_RP and EFI_MEMORY_RO if they're set?
>

I think the best strategy here is to simply apply the default
permissions first, and only then actually free the pages, clear them
etc etc That also fixes the existing issue where we may remap fewer
pages than what we actually freed. I'll add a patch to my series for
this.


>
> On 3/1/2023 1:57 PM, Ard Biesheuvel wrote:
> > On Wed, 1 Mar 2023 at 21:43, Taylor Beebe <t at taylorbeebe.com> wrote:
> >>
> >>
> >>
> >> On 2/11/2023 2:05 AM, Ard Biesheuvel wrote:
> >>> On Sat, 11 Feb 2023 at 01:56, Taylor Beebe <t at taylorbeebe.com> wrote:
> >>>>
> >>>> Hey Ard,
> >>>>
> >>>> Once the Memory Attribute Protocol is made available, Windows will have
> >>>> some expectations about its functionality. Can you run this test app
> >>>> created by me and Jiewen to ensure it meets the Windows requirements?
> >>>> Part of the test needed an AARCH64 implementation which I just added -
> >>>> let me know if it doesn't work.
> >>>>
> >>>
> >>> Thanks, this is rather helpful.
> >>>
> >>> There appears to be an issue related to
> >>> DEBUG_PROPERTY_DEBUG_CLEAR_MEMORY_ENABLED so I had to disable that to
> >>> run these tests, as otherwise, the DXE core tries to clear freed pages
> >>> before restoring the memory attributes.
> >>>
> >>> With that out of the way, the only test that fails is 'New
> >>> EfiLoaderCode buffer attributes expected' because this firmware build
> >>> maps loader code RWX, as existing boot stages for Linux are relying on
> >>> this (including the kernel itself at this point)
> >>
> >> It makes sense that the NewEfiLoaderCode test fails, but I am surprised
> >> the FreePagesWithProtectionAttributesTestCase passes. The test ensures
> >> that a page with EFI_MEMORY_RP and/or EFI_MEMORY_RO has those attributes
> >> cleared before attempting to free the page within the FreePage routine
> >> and is related to the concern Marvin had.
> >>
> >> Did you make a change to the core or is there an execution path I'm not
> >> seeing which allows that test to pass?
> >
> > No, I didn't make any additional changes to the core afair.
>


-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#101097): https://edk2.groups.io/g/devel/message/101097
Mute This Topic: https://groups.io/mt/96853149/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-